[selinkocalar]

The hard part with tools like this is maintaining context across different data models. GitHub PRs, Linear tickets, and LLM conversations all have different information architectures. Are you doing any semantic linking between related items, or just surface-level aggregation?

[wjsekfghks]

We use one model across all three surfaces called Task. Those do have different information architectures, i.e. Linear tickets have correspondence or comments, LLM conversations have chat history and code reviews have diffs and comments. But, at the end of the day, all those information are used to output the correct code and we are doing just that.